Skip to content

Benchmark

MLEB - Bar Exam QA

Embedding benchmarks measure how well models find related items, for example when matching a search to relevant results. Scores are percentages.

Models measured
18
Scores recorded
18
Sources
1
IQ weight
0.80

Score distribution

50.1%
Mean
53.4%
Median
9.3%
Std dev
38.6%
10th percentile
60.5%
90th percentile
34.7%
Min
34.7%66.2%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.80 for embedding models, measured over 18 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1voyage-4-largeBar Exam QA· MLEB data66.2%
2voyage-4Bar Exam QA· MLEB data62.7%
3Qwen3 Embedding 8BBar Exam QA· MLEB data60.5%
4Gemini Embedding 001Bar Exam QA· MLEB data57.8%
5Qwen3 Embedding 4BBar Exam QA· MLEB data56.6%
6voyage-4-liteBar Exam QA· MLEB data55.9%
7Qwen3 Embedding 0.6BBar Exam QA· MLEB data55%
8Jina Embeddings v5 Text SmallBar Exam QA· MLEB data54.4%
9Text Embedding 3 LargeBar Exam QA· MLEB data53.4%
10Jina Embeddings v5 Text NanoBar Exam QA· MLEB data53.3%
11Jina Embeddings v4Bar Exam QA· MLEB data46.9%
12snowflake-arctic-embed-l-v2.0Bar Exam QA· MLEB data45.8%
13snowflake-arctic-embed-m-v2.0Bar Exam QA· MLEB data45.3%
14Text Embedding 3 SmallBar Exam QA· MLEB data39.6%
15BGE M3Bar Exam QA· MLEB data39.1%
16Text Embedding Ada 002Bar Exam QA· MLEB data38.6%
17Multilingual E5 Large InstructBar Exam QA· MLEB data36.8%
18Granite Embedding Small English R2Bar Exam QA· MLEB data34.7%

Data credits: MLEB (Isaacus), MIT (licence)